Automatic tile position and orientation detection combining deep-learning and rule-based computer vision algorithms

Increasing interest in a tile-paving robot calls for a robust tile detection algorithm. This paper proposes the Ultra Clear Tile (UC-Tile) algorithm to detect corners and edges and assist tile paving automation in positioning and installation tasks. UC-Tile is designed to incorporate deep learning f...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Automation in construction Jg. 171; S. 106001
Hauptverfasser: Liu, Wenyao, Chen, Jinhua, Lyu, Zemin, Feng, Rui, Hu, Tong, Deng, Lu
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 01.03.2025
Schlagworte:
Construction robots
Deep learning
Intelligent construction
Segmentation
Tile paving
Deep learning
Construction robots
Segmentation
Tile paving
Intelligent construction
ISSN:0926-5805
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Increasing interest in a tile-paving robot calls for a robust tile detection algorithm. This paper proposes the Ultra Clear Tile (UC-Tile) algorithm to detect corners and edges and assist tile paving automation in positioning and installation tasks. UC-Tile is designed to incorporate deep learning for semantic segmentation with rule-based post-processing algorithms. The semantic segmentation algorithm investigated herein is a fine-tuned version of YOLOv8. UC-Tile mainly contributes to refitting the edges and locating the tile corners with tailored algorithms. A dataset is developed comprising 1486 images exhibiting varying patterns of tiles, captured under disparate heights and illumination conditions. Results indicate that UC-Tile outperforms common benchmark algorithms, and can achieve the highest mIoU 98.68 %, F1-Score 99.31 %, and the lowest 95-HD. Further, UC-Tile can accurately predict real distance and angles with small differences to ground truth values, thereby informing robotic movement control. This algorithm is expected to enable precise automatic tile paving. •UC-Tile algorithm is designed combining deep learning and rule-based algorithms.•UC-Tile can predict real distance and angle using 2D camera and point laser.•A tile dataset is developed comprising various variables in working conditions.•The performance of UC-Tile is evaluated comparing to eight benchmarks algorithms.•UC-Tile performs well in predicting tile segmentation, position and orientation.
ISSN:0926-5805
DOI:10.1016/j.autcon.2025.106001